Social Phenomena Simulation

mumpsimuspreviousΤεχνίτη Νοημοσύνη και Ρομποτική

25 Οκτ 2013 (πριν από 4 χρόνια και 2 μήνες)

67 εμφανίσεις

Social Phenomena Simulation


Paul Davidsson

Department of Computer Science, Malmö University, Malmö, Sweden

Harko Verhagen

Department of Computer and Systems Sciences, Stockholm University, Kista,
Sweden


Glossary

Agent (or software agent )

A self

contained entity that has a state and that is situated (able to perceive and
act) in an environment. In addition, agents are often assumed to be rational and
autonomous.


Cellular automaton

A mathematical structure modeling a set of cells that interact wi
th their neighbors.
Each cell has a set of neighbors and a state. All the cells update their values
simultaneously at discrete time steps. The new state of a cell is determined by the
current state of its neighbors according to a local function or rule.


M
icrolevel simulation

A type of simulation in which the specific behaviors of specific individuals are
explicitly modeled.



Definition of the Subject

Social phenomena simulation in the area of agent
-
based modeling and
simulation concerns the emulation of
the individual behavior of a group of social
entities, typically including their cognition, actions, and interaction. Agent
-
based
social simulation constitutes the intersection of three scientific fields, namely,
agent
-
based computing, the social sciences,

and computer simulation [
6
].
Agent
-
based computing is a research area mainly within computer science and
includes, e.

g., agent
-
based modeling, design, and programming. By the social
sciences we here refer to a large set of different sciences that study t
he
interaction among social entities, e.

g., social psychology, management science,
policy, and some areas of biology. Computer simulation concerns the study of
different techniques for simulating phenomena on a computer, e.

g., discrete
-
event, object

orie
nted, and equation

based simulation.


Introduction

Computer simulation consists of three main steps: (i) designing a model of an
actual or theoretical system, (ii) executing the model on a computer, and (iii)
analyzing the execution output. Already in the
early days of computer
development, simulation was used in different research areas to predict the
behavior of complex systems. Such simulations were typically based on
differential equations and focused on results at the aggregate level. These
models of,
for instance, predator

prey populations could result in fairly accurate
models but were limited in the sense that the models excluded individual
behavior and decision making, as well as interaction between individuals, and
were based on homogeneous agents.

The development of agent
-
based
modeling offers a possible solution to this problem with its (seemingly) natural
mapping onto interacting individuals with incomplete information and
capabilities, no global control, decentralized data, asynchronous computin
g, and
inclusion of heterogeneous agents. Agent
-
based simulation models also offer
the possibility of studying the dynamics of the interaction processes instead of
focusing on the (static) results of these processes [
16
,
26
].


Agent
-
based modeling can be tr
aced back to von Neumann, who in the 1950s
invented what was later termed
cellular automata
. These were used by Conway
in the 1970s when he constructed the well
-
known
Game of Life
. It is based on
very simple rules determining the life and death of the cell
s in a virtual world in
the form of a 2

D grid. Inspired by this work, researchers developed more
-
refined models, often modeling the social behavior of groups of animals or
artificial creatures. One example is the Boid model by Reynolds [
24
], which
simulat
es coordinated animal motion such as bird flocks and fish schools. With
respect to human societies, Epstein and Axtell [
8
] developed in the 1990s one
of the first agent
-
based models, called Sugarscape, to explore the role of social
phenomena such as season
al migrations, pollution, sexual reproduction,
combat, and transmission of disease. This work is in spirit closely related to one
of the best
-
known and earliest examples of the use of simulation in social
science, namely, the Schelling model [
27
], in which

cellular automata were
used to simulate the emergence of segregation patterns in neighborhoods based
on a few simple rules expressing the preferences of the agents. Another pioneer
from the 1950s worth mentioning is Barricelli [
2
], who to some extent used

agent
-
based modeling for simulating biological systems.


The cellular automata models closely resemble the models used in statistical
physics, which has inspired physicists to include the simulation of social
phenomena in large
-
scale social systems in the
ir research agenda. In this area,
sometimes referred to as sociophysics, phenomena such as opinion spreading in
a society and competition between languages have been studied. These models
originally described the behavior of atoms and molecules, which are
quite
simple objects, and the macrolevel phenomena caused by their interaction
(rather than by complex behavior of the individual as in the case of humans).
Thus, in these models little attention is paid to individual variation and the
individual decision
-
making is rather primitively modeled. A prominent example
of sociophysics is the work of Galam [
10
].


To sum up, we can identify two main approaches to social simulation:



Macrolevel (or equation

based) simulation, which is typically based on
mathematical m
odels. It views the set of individuals (the population) as a
structure that can be characterized by a number of variables.



Microlevel (or agent
-
based) simulation, in which the specific behaviors of
specific individuals are explicitly modeled. In contrast t
o macrolevel
simulation, it views the structure as emerging from the interactions between
individuals and thus exploring the standpoint that complex effects need not
have complex causes.

As argued by Van Parunak et al. [
21
], agent
-
based modeling is most
ap
propriate for domains characterized by a high degree of localization and
distribution and dominated by discrete decision. Equation

based modeling, on
the other hand, is most naturally applied to systems that can be modeled
centrally and in which the dynami
cs are dominated by physical laws rather than
information processing. We will here focus on agent
-
based models, particularly
those that have a richer representation of the individual than the cellular
automata and statistical physics models.


Why Simulate
Social Phenomena?

Simulation of social phenomena can be done for different purposes, e.

g.,



Supporting social
-
theory building;



Supporting the engineering of systems, e.

g., validation, testing, etc.;



Supporting planning, policy making, and other decision
making;



Training, in order to improve a person's skills in a certain domain.


It is possible to distinguish between four types of end users:
scientists
, who use
social phenomena simulation in the research process to gain new knowledge,
policymakers
, who us
e it for making strategic decisions,
managers

(of systems),
who use it to make operational decisions, and
other professionals
, such as
architects, who use it in their daily work. We will now describe how these types
of end users may use simulation of socia
l phenomena for different purposes.


Supporting Social
-
Theory Building

In the context of social
-
theory building, agent
-
based simulation can be seen as
an experimental method or as theories in themselves [
26
]. In the former case,
simulations are run to test

the predictions of theories, whereas in the latter case
simulations in themselves are formal models of theories. Formalizing the
ambiguous, natural
-
language

based theories of the social sciences helps to find
inconsistencies and other problems, and thus c
ontributes to theory building.


Using agent
-
based simulation studies as an experimental tool offers great
possibilities. Many experiments with human societies are either unethical or
even impossible to conduct. Experiments in silico, on the other hand, are

fully
possible. These can also breathe new life into the ever

present debate in
sociology on the micro
-
macro link [
1
]. Agent
-
based models mostly focus on the
emergence of macrolevel properties from the local interaction of adaptive
agents that influence o
ne another [
17
,
26
]. However, simulations in
computational organization theory [
4
,
22
], for example, often try to analyze the
influence of macrolevel

phenomena on individuals. Using agent
-
based models to
simulate the bidirectional relation between micro
-

and macrolevel concepts
would provide tools to analyze the theoretical consequences of the work done
by theorists such as Habermas, Giddens, and Bourd
ieu, to name a few [
26
].


Supporting the Engineering of Systems

Many new technical systems are distributed and involve complex interactions
between humans and machines. The properties of agent
-
based simulation make
it especially suitable for simulating the
se kinds of systems. The idea is to model
the behavior of human users in terms of software agents. In particular, this
seems useful in situations where it is too expensive, difficult, inconvenient,
tiresome, or even impossible for real human users to test
out a new technical
system. Of course, also the technical system, or parts thereof, may be
simulated. For instance, if the technical system includes hardware that is
expensive and/or special purpose, it is natural to simulate also this part of the
system w
hen testing out the control software. An example of such a case is the
testing of control systems for "intelligent buildings," where agents simulate the
behavior of the people in the building [
5
].


Supporting Planning, Policy Making, and Other Decision Mak
ing

Here the focus is on exploring different possible future scenarios in order to
choose between alternative actions. Besides this type of prediction, simulation
of social phenomena may be used for analysis, i.

e., to gain deeper knowledge
and understandi
ng of a certain phenomenon.


An area in which several studies of this kind have been carried out is disaster
management, such as experiments concerning different roles and the efficiency
of reactions to emergencies [
18
]. Based on individuals' observations,

personal
characteristics and skills, past experience and role characteristics, and social
network, the agents create a plan to execute. Each agent represents a human
being (acting in a particular role). The effect of adding a role (floor warden) in a
fire

alarm scenario upon the evacuation efficiency in an abstract environment is
analyzed. In another approach, the agents are placed in an environment based
on GIS (geographical information system) data, thereby tying the simulation
closer to the physical rea
lity [
29
]. In yet another study, real
-
world data were
used for both the environment and the agents' internal decision
-
making model
to analyze the effect of different insurance policies on the willingness of agents
to pay for a disaster insurance policy [
3
]
.


Another application area for this type of simulation study is disease spreading.
Typically, agents are used to represent human beings and the simulation model
is linked to real
-
world geographical data. One study [
32
] also included agents
that represent
towns acting as the epicenter of disease outbreak. The town
agent's behavior repertoire consisted of different containment strategies. The
simulation model can be quickly adapted to local circumstances via the
geographical data (given that there is data on

the population as well) and is
used to determine the effects of different containment strategies.


A third area where agent
-
based social simulation has been used to support
planning and policy
-
making is traffic and transport. An example of this is the
simulation of all car travel in Switzerland during morning peak traffic [
23
].


Training

The main advantage of using simulation for training purposes is to be part of a
real
-
world
-
like situation without real
-
world consequences. Especially in the
military th
e use of simulation for training purposes is widespread. Also in
medicine, where mistakes can be very expensive in terms of money and lives,
the use of simulation in education is on the rise.


An early product in this area was a tool to help train police o
fficers to manage
large public gatherings such as crowds, demonstrations, and marches [
31
].
Another early example of agent
-
based simulation for training purposes is Steve
[
19
,
25
]. Steve was an agent integrated with voice synthesis software and
virtual real
ity software providing a very realistic training environment for
controlling the engine room of a virtual US Navy surface ship.


An example of a more recent project is the PSI agent [
15
]. Whereas in most
cases the simulator training is aimed at training pr
actical skills or decision
-
making, this work focuses on acquiring theoretical insights in the realm of
psychological theory. The simulation enables students to explore psychological
processes without ethical problems.


Simulating Social Phenomena

One of the first, and most simple, way of performing microlevel simulation is often
called
dynamic microsimulation

[
11
,
12
]. It is used to simulate the effect of the
passing of time on individuals. Data from a (preferably large) random sample from
the popul
ation to be simulated is used to initially characterize the simulated
individuals. Some examples of sampled features are: age, sex, employment status,
income, and health status. A set of transition probabilities are used to describe how
these features will

change over a given time period, e.

g., there is a probability that
an employed person will become unemployed over the course of a year. The
transition probabilities are applied to the population for each individual in turn and
then repeatedly reapplied f
or a number of simulated time periods. Sometimes it is
necessary to also model changes in the population, e.

g., birth, death, and
marriage. This type of simulation can be used to, e.

g., predict the outcome of
different social policies. However, the quali
ty of such simulations depends on the
quality of:



the random sample, which must be representative, and



the transition probabilities, which must be valid and complete.


In traditional microsimulation, the behavior of each individual is regarded as a
"black
box." The behavior is modeled in terms of probabilities and no attempt is
made to justify these in terms of individual preferences, decisions, plans, etc.
Also, each simulated individual is considered in isolation without regard to their
interaction with o
thers. Thus, better results may be gained if cognitive
processes and communication between individuals are also simulated.


Opening the black box of individual decision making can be done in several
ways. The first layer to add is often individual psycholo
gy; for instance, the so

called beliefs, desires, and intentions (BDI) model is often used. Models of
individual cognition used in agent
-
based social simulation include the use of
Soar (a computer implementation of Allen Newell's unified theory of cognitio
n
[
20
]), which was used in Steve (discussed in Sect. "Why Simulate Social
Phenomena?").


For the simulation of social behavior the agents need to be equipped with
mechanisms for reasoning at the social level (unless the social level is regarded
as emerging

from individual behavior and decision making). Several models
have been based on theories from economics, social psychology, sociology, etc.
An example of this is provided by Guye

Vuillème [
13
], who has developed an
agent
-
based model for simulating human
interaction in a virtual
-
reality
environment. The model is based on sociological concepts such as roles, values,
and norms and motivational theories from social psychology to simulate persons
with social identities and relationships. Another example is the

Consumat model
[
14
], a metamodel combining several psychological theories on decision making
in a consumer situation, used, for instance, to investigate different flood
-
management policies [
3
]. Also, nonsymbolic approaches such as neural
networks have bee
n used to model agents' decision making [
18
].


Future Directions

In a recent study of applications of agent
-
based simulation [
7
], it was concluded
that even if agent
-
based simulation seems a promising approach to many
problems involving the simulation of
complex systems of interacting entities
such as social phenomena, it seems that the full potential of the agent concept
often is not utilized. For instance, most models have very primitive agent
cognition, in particular if the number of agents involved is
large.


Regarding future applications, Fiedrich and Burghardt [
9
] argue that agent
-
based simulation is a very promising approach to disaster management
practice. In particular, agent
-
based social simulation in combination with
sophisticated visualization techniques, such as virtual reality, in the form of
"serious game
s ," has the potential to provide very powerful training
environments. In the context of military training, Stone [
28
] provides some
interesting applications.


Further Reading

Classic books in the area of simulation of social behavior include "Growing
Arti
ficial Societies: Social Science from the Bottom Up" by Epstein and Axtell [
8
]
and "Simulation for the Social Scientist" by Gilbert and Troitzsch [
12
].


More recent findings can be found in, e.

g., the Journal of Artificial Societies and
Social Simulation
(
http://jasss.soc.surrey.ac.uk
) and the proceedings of, e.

g.,
the International Workshop series on Multi
-
Agent
-
Based Simulation (MABS)
(
http://www.pcs.usp.br/~mabs
/
), the World Congress in Social Simulation
(WCSS) [
30
], the conference of the European Social Simulation Association
(ESSA) (
http://www.essa.eu.org/
), and the series of Agent Workshops in
Chicago (
http://www.agent2005.anl.gov/
).


Bibliography

1.

Alexander JC, Giesen B, Münch R, Smelser NJ (eds) (1987) The Micro
-
Macro
Link. University of California Press, Berkeley

2.

Barricelli NA (1957) Symbiogenetic

evolution processes realized by artificial
methods. Methodos 9(35
-
36):143
-
182

3.

Brouwers L, Verhagen H (2004) Applying the Consumat model to flood
management policies. In: Müller J, Seidel M (eds) 4th Workshop on Agent
-
Based
Simulation. SCS, Montpellier, p
p 29
-
34

4.

Carley KM, Prietula M (eds) (1994) Computational Organization Theory. Erlbaum,
Hillsdale

5.

Davidsson P (2000) Multi
-
agent
-
based simulation: beyond social simulation. In:
Moss S, Davidsson P (eds) Multi
-
Agent
-
Based Simulation. Lecture Notes in
Compu
ter Science, vol 1979. Springer, Berlin, pp 98
-
107

6.

Davidsson P (2002) Agent
-
based social simulation: a computer science view. J
Artif Soc Soc Simul 5(1)

7.

Davidsson P, Holmgren J, Kyhlbäck H, Mengistu D, Persson M (2007)
Applications of multi
-
agent
-
based s
imulation. In: Antunes L, Takadama K (eds)
Multi
-
Agent
-
Based Simulation VII. Lecture Notes in Computer Science, vol 4442.
Springer, Berlin

8.

Epstein JM, Axtell RL (1996) Growing Artificial Societies: Social Science from the
Bottom Up. MIT Press, Cambridge

9.

Fiedrich F, Burghardt P (2007) Emergency response information systems:
emerging trends and technologies: agent
-
based systems for disaster
management. Commun ACM 50(3):41
-
42

10.

Galam S (2008) Sociophysics: a review of Galam models. arXiv:0803.1800
http://arxiv.org/abs/0803.1800
. Accessed 06 Aug 2008

11.

Gilbert N (1994) Computer Simulation of Social Processes. Social Research
Update, Issue 6, Department of Sociology, University of Surrey, UK.
http://www.soc.surrey.ac.uk/sru/
. Accessed 06 Aug 2008

12.

Gilbert N, Troitzsch KG (2005) Simulation for the Social Scientist, 2nd edn.
Open University Press, Maidenhead

13.

Guye
-
Vuillème A (2004) Simulation of nonverbal social interacti
on and small
groups dynamics in virtual environments. Ph

D thesis, Ècole Polytechnique
Fédérale de Lausanne, No 2933

14.

Janssen MA, Jager W (1999) An Integrated Approach to Simulating Behavioural

Processes: A Case Study of the Lock
-
in of Consumption Patterns. J Artif Soc Soc
Simul 2(2)

15.

Künzel J, Hämmer V (2006) Simulation in university education: the artificial
agent PSI as a teaching tool. Simulation 82(11):761
-
768

16.

Lansing JS (2002) "Artificial

societies" and the social sciences. Artif Life 8:279
-
292

17.

Macy MW, Willer R (2002) From factors to actors: computational sociology and
agent
-
based modeling. Annu Rev Sociol 28:143
-
166

18.

Massaguer D, Balasubramanian V, Mehrotra S, Venkatasubramanian N (2006
)
Multi
-
agent simulation of disaster response. In: Proceedings of the 1st
International Workshop on Agent Technology for Disaster Management 2006.
Hakodate

19.

Méndez G, Rickel J, de Antonio A (2003) Steve meets Jack: the integration of
an intelligent tutor a
nd a virtual environment with planning capabilities. In:
Intelligent Virtual Agents. Lecture Notes on Artificial Intelligence, vol 2792.
Springer, Berlin, pp 325
-
332

20.

Newell A (1994) Unified Theories of Cognition. Harvard University Press,
Cambridge

21.

Parun
ak HVD, Savit R, Riolo RL (1998) Agent
-
based modeling vs. equation

based modeling: a case study and users' guide. In: Sichman JS, Conte R, Gilbert
N (eds) Multi
-
Agent Systems and Agent
-
Based Simulation. Lecture Notes in
Computer Science, vol 1534. Springer
, Berlin, pp 10
-
26

22.

Prietula MJ, Carley KM, Gasser L (eds) (1998) Simulating Organizations:
Computational Models of Institutions and Groups. MIT Press, Cambridge

23.

Raney B, Cetin N, Völlmy A, Vrtic M, Axhausen K, Nagel K (2003) An Agent
-
Based Microsimulatio
n Model of Swiss Travel: First Results. J Netw Spatial Econ
3(1):23
-
41

24.

Reynolds CW (1987) Flocks, herds, and schools: a distributed behavioral model.
Comput Graph 21(4):25
-
34 (SIGGRAPH '87 Conference Proceedings)

25.

Rickel

J, Lewis Johnson W (1999) Animated agents for procedural training in
virtual reality: perception, cognition, and motor control. Appl Artif Intell 13:343
-
382

26.

Sawyer RK (2003) Artificial societies
-

multi
-
agent systems and the micro
-
macro link in sociologi
cal theory. Sociol Meth Res 31(3):325
-
363

27.

Schelling TC (1971) Dynamic models of segregation. J Math Sociol 1(1):143
-
186

28.

Stone B (2005) Serious gaming. Def Manag J (31):142
-
144

29.

Takahashi T (2007) Agent based disaster simulation evaluation and its
probabi
lity model interpretation. In: Proceedings of the 4th International
Conference on Intelligent Human

Computer Systems for Crisis Response and
Management, 2007, Deltft, pp 369
-
376

30.

Takahashi S, Sallach D, Rouchier J (2007) Advancing Social Simulation
-

The
F
irst World Congress. Springer, Berlin

31.

Williams R (1993) An agent based simulation environment for public order
management training. In: Western Simulation Multiconference, Object

Oriented
Simulation Conference (OOS '93), San Diego 151
-
156

32.

Yergens D, Hine
r J, Denzinger J, Noseworthy T (2006) IDESS
-

A multi
-
agent
-
based simulation system for rapid development of infectious disease models. Int
Trans Syst Sci Appl 1(1):51
-
58